Cautionary tails

Phase III randomised controlled trials (RCTs) are usually analysed and published according to a statistical plan, when the number of events is expected be sufficient establish (at P < 0.05) or exclude (with power 80% 90%) prespecified difference in primary time-to-event outcome. For that evaluate treatments for metastatic cancer, end point most often overall survival (OS) progression-free (PFS), whereas adjuvant therapy early disease it OS relapse-free (RFS). Survival data presented as Kaplan–Meier curves, which allow inclusion participants recruited over time with variable follow-up, assumption treatment will influence equally early- late-occurring events. The statistic given such reports hazard ratio (HR)—the any short interval between experimental control arms—with its confidence value, aid characterising measured HR unity (representing no effect). Use assumes proportional hazards (i.e. relative rate arms remains constant during follow-up), not satisfied even evaluated.1Rulli E. Ghilotti F. Biagioli et al.Assessment aggregate data: systematic review on methodology clinical using endpoint.Br J Cancer. 2018; 119: 1456-1463Crossref PubMed Scopus (16) Google Scholar,2Rahman R. Fell G. Ventz S. al.Deviation from randomized phase 3 oncology: prevalence, associated factors, implications.Clin Cancer Res. 2019; 25: 6339-6345Crossref (25) Scholar ‘positive’ trial where 1, there separation curves treatments, impressive visual differences occurring at later i.e. ‘tails’ curves. In light results revealing durable benefits immune checkpoint inhibitors some cancers,3Harris S.J. Brown J. Lopez Yap T.A. Immuno-oncology combinations: raising tail curve.Cancer Biol Med. 2016; 13: 171-193Crossref (63) have been recent calls shift focus away median towards outcomes tail-end curves.4Hellmann M.D. Kris M.G. Rudin C.M. Medians milestones describing path cancer cures: telling “tails”.JAMA Oncol. 2: 167-168Crossref (31) However, due limited curve least well defined. Since about 2005, high-impact journals publishing RCTs required patients ‘at risk’—subjects event interest still possible specified time—be below Despite this, readers may fail notice how few define tails hence unreliable interpretation these points. Ideally, should updated show mature data. Given continuous patient recruitment study period, (albeit prespecified) times analysed, invariably transform time, changes both late numbers risk. Data maturity implies followed long enough assessed, so based an adequate incurable advanced cancers, high proportion subjects progressed died. Trials majority those destined relapse done so. study, however, we were able locate follow-up only 20% evaluating breast, lung prostate cancer.5Elimova Moignard Li X. al.Updating cancer.JAMA 2020; (forthcoming)Google ability oncologists examine appears rare. To illustrate problem, solution, three high-profile Figure 1. CheckMate 066 418 untreated melanoma without BRAF mutation nivolumab dacarbazine. was OS. First 2015,6Robert C. Long G.V. Brady B. al.Nivolumab previously mutation.N Engl 2015; 372: 320-330Crossref (3580) reported 0.42 (P 0.001); shown out 16 months, but 45 22 risk beyond 1 year, poorly defined (Figure 1A). Longer-term follow up 2019,7Ascierto P.A. Robert al.Survival wild-type treated therapy. Three-year trial.JAMA 5: 187-194Crossref (149) plotted 4 years consistent benefit 1B). CLEOPATRA 808 women HER2+ breast docetaxel trastuzumab either pertuzumab placebo. Primary secondary points PFS OS, respectively. 2012 analysis carried 10 months after ended.8Baselga Cortes Kim S.-B. al.Pertuzumab plus cancer.N 2012; 366: 109-119Crossref (1746) HRs 0.62 0.64, respectively (both highly significant). 39 defining their 1C). Increasingly 2013, 2015 2020, showing long-term 1D), >8 publication.9Swain S.M. Miles D. S.B. al.Pertuzumab, trastuzumab, HER2-positive (CLEOPATRA): end-of-study double-blind, randomised, placebo-controlled, study.Lancet 21: 519-530Abstract Full Text PDF (125) Three subsequent publications might seem excessive, this trial, along 066, practice changing, assessment toxicity paramount. contrast, CALGB 49907 compared standard chemotherapy (doxorubicin cyclophosphamide cyclophosphamide, methotrexate 5-fluorouracil) single-agent capecitabine 633 older cancer. This 2009, 29 months.10Muss H.B. Berry D.A. Cirrincione C.T. al.Adjuvant early-stage 2009; 360: 2055-2065Crossref (407) RFS 5 years, respectively, showed significantly better (HR = 0.48, 0.001 RFS; 0.54, 0.02 OS). separated, relapse, and, 15 death 1E). Ten-year 2019,11Muss Polley M.-Y.C. al.Randomized regimens versus cancer: 10-year update trial.J Clin 37: 2338-2348Crossref evident 1F), 0.84 0.16). Although two predictive results, strong bias investigators pharmaceutical sponsors positive maintained enhanced, against reporting not.12Krzyzanowska M.K. Pintilie M. Tannock I.F. Factors failure publish large oncology meeting.JAMA. 2003; 290: 495-501Crossref (264) published,5Elimova uncertainty whether separated do reflect term benefit. Tails viewed caution.